[
https://issues.apache.org/jira/browse/CASSANDRA-9457?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14558793#comment-14558793
]
Sylvain Lebresne commented on CASSANDRA-9457:
---------------------------------------------
I'm a bit uncomfortable special casing certain types like that, but I don't
have a much better solution to suggest so I'm fine with it. However, we at
least need to accept empty values for "custom" type, so instead of having a
whitelist of types for which empty values do make sense, I'd prefer adding a
blacklist of types that do accept them but for which it doesn't really make
sense. In particular such blacklist won't ever have to be updated since from
now on, when we add new types for which empty values don't make sense, we
actually don't allow them in the first place, so empty value will be rejected
by type-checking for these new types.
I would also have a preference for adding some
{{acceptEmptyValuesEvenThoughItDoesntReallyMakeSense()}} method to
{{AbstractType}} rather hardcoding some list.
> Empty INITCOND treated as null in aggregate
> -------------------------------------------
>
> Key: CASSANDRA-9457
> URL: https://issues.apache.org/jira/browse/CASSANDRA-9457
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Reporter: Olivier Michallat
> Assignee: Robert Stupp
> Priority: Minor
> Fix For: 2.2.0 rc1
>
>
> Given the following test data:
> {code}
> cqlsh:test> create table foo(k int, v int, primary key(k,v));
> cqlsh:test> insert into foo(k,v) values(1,1);
> cqlsh:test> insert into foo(k,v) values(1,2);
> cqlsh:test> insert into foo(k,v) values(1,3);
> {code}
> And the following aggregate definition:
> {code}
> cqlsh:test> CREATE FUNCTION cat(s text, v int)
> RETURNS NULL ON NULL INPUT
> RETURNS text
> LANGUAGE java
> AS 'return s + v;';
> cqlsh:test> CREATE AGGREGATE cats(int) SFUNC cat STYPE text INITCOND '';
> {code}
> The following should return '123', but it returns null:
> {code}
> cqlsh:test> select cats(v) from foo where k = 1;
> test.cats(v)
> ---------------
> {code}
> The empty INITCOND is treated as null, and the SFUNC is never called.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
